Summary
Aaron's directive (2026-09-25): the meta self-improvement — recording, refining, and curating skills and vault knowledge — must happen ALONGSIDE research and development, as part of both mode protocols, not after them. And research loops must emit plotted pulses/data at every record boundary (reports or inline widgets).
What to encode
Research mode (packages/extension/modes/research/)
pack.toml — new gate in the analyze phase, record-boundary-curation (kind: derived, owner: director): every record boundary emits the loop's visual record (per-run pulse/data plot artifacts + the dashboard widget refresh) AND commits the curation delta — skill findings filed, vault knowledge cards for measured laws, proposals' status. Meta-improvement is a first-class loop output.SKILL.md — extend the loop's record step with the plotting ritual + the curation items (they are currently implicit in the analyzer's skill-friction duty; make them explicit and reported).
Develop mode (packages/extension/modes/develop/)
pack.toml — new gate in the integrate phase, meta-improvement-delta (kind: derived, owner: director): each dev walk records its own meta-delta — skill friction encountered during implementation, doctrine/docs updated alongside the code, and the hypothesis seed for research at close.SKILL.md — matching extension to the dev walk's record step.
Evidence (why now)
The pasqal-gate-autoresearch campaign (6 loops) validated the pattern organically: 18+ skill-friction findings (armonissima#108 + supplements), 2 stale problem-card lessons corrected, a measured waveform-reconstruction doctrine vaulted (reference_rydberg_waveform_reconstruction.md), a live per-loop pulse widget (pasqal-cz-pulse-board), and 12 skill proposals — all as byproducts of the research loop. The directive makes them protocol, not byproducts.
Acceptance
Summary
Aaron's directive (2026-09-25): the meta self-improvement — recording, refining, and curating skills and vault knowledge — must happen ALONGSIDE research and development, as part of both mode protocols, not after them. And research loops must emit plotted pulses/data at every record boundary (reports or inline widgets).
What to encode
Research mode (
packages/extension/modes/research/)pack.toml— new gate in theanalyzephase,record-boundary-curation(kind: derived, owner: director): every record boundary emits the loop's visual record (per-run pulse/data plot artifacts + the dashboard widget refresh) AND commits the curation delta — skill findings filed, vault knowledge cards for measured laws, proposals' status. Meta-improvement is a first-class loop output.SKILL.md— extend the loop's record step with the plotting ritual + the curation items (they are currently implicit in the analyzer's skill-friction duty; make them explicit and reported).Develop mode (
packages/extension/modes/develop/)pack.toml— new gate in theintegratephase,meta-improvement-delta(kind: derived, owner: director): each dev walk records its own meta-delta — skill friction encountered during implementation, doctrine/docs updated alongside the code, and the hypothesis seed for research at close.SKILL.md— matching extension to the dev walk's record step.Evidence (why now)
The pasqal-gate-autoresearch campaign (6 loops) validated the pattern organically: 18+ skill-friction findings (armonissima#108 + supplements), 2 stale problem-card lessons corrected, a measured waveform-reconstruction doctrine vaulted (
reference_rydberg_waveform_reconstruction.md), a live per-loop pulse widget (pasqal-cz-pulse-board), and 12 skill proposals — all as byproducts of the research loop. The directive makes them protocol, not byproducts.Acceptance